Machine Learning for Red Teams, Part 1

SpecterOps maps where AI helps offensive security (recon, tooling, vuln discovery) and where it still fails, stressing that results hinge on the harness: context, tools, state, independent validation, and human judgment.

Provides verifiable, transparent evidence of advanced LLMs performing complex, multi-step binary and browser exploitation autonomously under strict ASLR and memory randomization defenses.

Azure Machine Learning stores Jupyter notebooks in Storage Accounts; attackers with write access can modify them to gain code execution without interaction. NetSPI also released a credential dumper.

NetSPI demonstrates obfuscating HTML smuggling with Web Assembly, embedding payloads in Wasm so files reassemble victim-side and evade network detections that flag suspicious outbound file downloads.

NetSPI's new AI/ML pentesting solution offers ML security assessments for models including LLMs, an infrastructure security assessment, and web app testing for LLM-integrated functionality.

Using the PyTorch-nightly compromise as a case study, this post explores attacking AI via its supply chain: typosquatted dependencies, credential theft from build machines, and model integrity risks.